set -e
-# bug 2732 2986
-ALWAYS_EXCEPT="17 20b"
+# 20b: bug 2986
+ALWAYS_EXCEPT=" 20b"
LUSTRE=${LUSTRE:-`dirname $0`/..}
-UPCALL=${UPCALL:-$PWD/recovery-small-upcall.sh}
+
. $LUSTRE/tests/test-framework.sh
init_test_env $@
-. ${CONFIG:=$LUSTRE/tests/cfg/local.sh}
+. ${CONFIG:=$LUSTRE/tests/cfg/lmv.sh}
build_test_filter
+assert_env MDSCOUNT
# Allow us to override the setup if we already have a mounted system by
# setting SETUP=" " and CLEANUP=" "
SETUP=${SETUP:-"setup"}
CLEANUP=${CLEANUP:-"cleanup"}
-
-make_config() {
+gen_config() {
rm -f $XMLCONFIG
- add_mds mds --dev $MDSDEV --size $MDSSIZE
- add_lov lov1 mds --stripe_sz $STRIPE_BYTES\
- --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
+
+ if [ "$MDSCOUNT" -gt 1 ]; then
+ add_lmv lmv1_svc
+ for mds in `mds_list`; do
+ MDSDEV=$TMP/${mds}-`hostname`
+ add_mds $mds --dev $MDSDEV --size $MDSSIZE --lmv lmv1_svc
+ done
+ add_lov_to_lmv lov1 lmv1_svc --stripe_sz $STRIPE_BYTES \
+ --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
+ MDS=lmv1
+ else
+ add_mds mds1 --dev $MDSDEV --size $MDSSIZE
+ add_lov lov1 mds1 --stripe_sz $STRIPE_BYTES \
+ --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
+ MDS=mds1
+
+ fi
+
add_ost ost --lov lov1 --dev $OSTDEV --size $OSTSIZE
add_ost ost2 --lov lov1 --dev ${OSTDEV}-2 --size $OSTSIZE
- add_client client mds --lov lov1 --path $MOUNT
+ add_client client ${MDS} --lov lov1 --path $MOUNT
}
setup() {
- make_config
+ gen_config
+ start_krb5_kdc || exit 1
+ start_lsvcgssd || exit 2
+ start_lgssd || exit 3
start ost --reformat $OSTLCONFARGS
start ost2 --reformat $OSTLCONFARGS
[ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
- start mds $MDSLCONFARGS --reformat
- zconf_mount `hostname` $MOUNT
+ for mds in `mds_list`; do
+ start $mds --reformat $MDSLCONFARGS
+ done
+ grep " $MOUNT " /proc/mounts || zconf_mount `hostname` $MOUNT
}
cleanup() {
zconf_umount `hostname` $MOUNT
- stop mds ${FORCE} $MDSLCONFARGS
+ for mds in `mds_list`; do
+ stop $mds ${FORCE} $MDSLCONFARGS
+ done
stop ost2 ${FORCE} --dump cleanup.log
stop ost ${FORCE} --dump cleanup.log
-}
-
-replay() {
- do_mds "sync"
- do_mds 'echo -e "device \$mds1\\nprobe\\nnotransno\\nreadonly" | lctl'
- do_client "$1" &
- shutdown_mds -f
- start_mds
- wait
- do_client "df -h $MOUNT" # trigger failover, if we haven't already
+ stop_lgssd
+ stop_lsvcgssd
}
if [ ! -z "$EVAL" ]; then
exit $?
fi
+if [ "$ONLY" == "setup" ]; then
+ setup
+ exit
+fi
+
if [ "$ONLY" == "cleanup" ]; then
sysctl -w portals.debug=0 || true
cleanup
REFORMAT=--reformat $SETUP
unset REFORMAT
+[ "$ONLY" == "setup" ] && exit
+
test_1() {
drop_request "mcreate $MOUNT/1" || return 1
drop_reint_reply "mcreate $MOUNT/2" || return 2
#define OBD_FAIL_MDS_CLOSE_NET 0x115
sleep 2
kill -USR1 $PID
+ cancel_lru_locks MDC # force the close
echo "waiting for multiop $PID"
wait $PID || return 2
do_facet client munlink $MOUNT/$tfile || return 3
# Bug 113, check that readdir lost recv timeout works.
test_13() {
- mkdir /mnt/lustre/readdir
- touch /mnt/lustre/readdir/newentry
+ mkdir /mnt/lustre/readdir || return 1
+ touch /mnt/lustre/readdir/newentry || return
# OBD_FAIL_MDS_READPAGE_NET|OBD_FAIL_ONCE
do_facet mds "sysctl -w lustre.fail_loc=0x80000104"
- ls /mnt/lustre/readdir || return 1
+ ls /mnt/lustre/readdir || return 3
do_facet mds "sysctl -w lustre.fail_loc=0"
- rm -rf /mnt/lustre/readdir
+ rm -rf /mnt/lustre/readdir || return 4
}
run_test 13 "mdc_readpage restart test (bug 1138)"
}
run_test 15 "failed open (-ENOMEM)"
+stop_read_ahead() {
+ for f in /proc/fs/lustre/llite/*/read_ahead; do
+ echo 0 > $f
+ done
+}
+
+start_read_ahead() {
+ for f in /proc/fs/lustre/llite/*/read_ahead; do
+ echo 1 > $f
+ done
+}
+
+# recovery timeout. This actually should be taken from
+# obd_timeout
+RECOV_TIMEOUT=30
+
test_16() {
do_facet client cp /etc/termcap $MOUNT
sync
+ stop_read_ahead
#define OBD_FAIL_PTLRPC_BULK_PUT_NET 0x504 | OBD_FAIL_ONCE
sysctl -w lustre.fail_loc=0x80000504
do_facet client "cmp /etc/termcap $MOUNT/termcap" && return 1
sysctl -w lustre.fail_loc=0
# give recovery a chance to finish (shouldn't take long)
- sleep 1
+ sleep $RECOV_TIMEOUT
do_facet client "cmp /etc/termcap $MOUNT/termcap" || return 2
+ start_read_ahead
}
run_test 16 "timeout bulk put, evict client (2732)"
test_17() {
-#define OBD_FAIL_PTLRPC_BULK_GET_NET 0x0503 | OBD_FAIL_ONCE
- # will get evicted here
+ # OBD_FAIL_PTLRPC_BULK_GET_NET 0x0503 | OBD_FAIL_ONCE
+ # client will get evicted here
sysctl -w lustre.fail_loc=0x80000503
- do_facet client cp /etc/termcap $MOUNT && return 1
+ do_facet client cp /etc/termcap $DIR/$tfile
- do_facet client "cmp /etc/termcap $MOUNT/termcap" && return 1
+ sleep $RECOV_TIMEOUT
sysctl -w lustre.fail_loc=0
- do_facet client "cmp /etc/termcap $MOUNT/termcap" || return 2
+ do_facet client "df $DIR"
+ # expect cmp to fail
+ do_facet client "cmp /etc/termcap $DIR/$tfile" && return 1
+ do_facet client "rm $DIR/$tfile" || return 2
+ return 0
}
run_test 17 "timeout bulk get, evict client (2732)"
drop_ldlm_cancel "chmod 0777 $f" || echo evicted
do_facet client checkstat -v -p 0777 $f || echo evicted
+ # let the client reconnect
+ sleep 5
do_facet client "munlink $f"
}
run_test 19a "test expired_lock_main on mds (2867)"
mkdir -p $DIR/$tdir
multiop $DIR/$tdir/${tfile} O_wc &
MULTI_PID=$!
- usleep 500
+ sleep 1
cancel_lru_locks OSC
#define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
do_facet ost sysctl -w lustre.fail_loc=0x80000308
}
run_test 20b "ldlm_handle_enqueue error (should return error)"
+test_21a() {
+ mkdir -p $DIR/$tdir-1
+ mkdir -p $DIR/$tdir-2
+ multiop $DIR/$tdir-1/f O_c &
+ close_pid=$!
+
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000129"
+ multiop $DIR/$tdir-2/f Oc &
+ open_pid=$!
+ sleep 1
+ do_facet mds "sysctl -w lustre.fail_loc=0"
+
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000115"
+ kill -USR1 $close_pid
+ cancel_lru_locks MDC # force the close
+ wait $close_pid || return 1
+ wait $open_pid || return 2
+ do_facet mds "sysctl -w lustre.fail_loc=0"
+
+ $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
+ $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
+
+ rm -rf $DIR/$tdir-*
+}
+run_test 21a "drop close request while close and open are both in flight"
+
+test_21b() {
+ mkdir -p $DIR/$tdir-1
+ mkdir -p $DIR/$tdir-2
+ multiop $DIR/$tdir-1/f O_c &
+ close_pid=$!
+
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000107"
+ mcreate $DIR/$tdir-2/f &
+ open_pid=$!
+ sleep 1
+ do_facet mds "sysctl -w lustre.fail_loc=0"
+
+ kill -USR1 $close_pid
+ cancel_lru_locks MDC # force the close
+ wait $close_pid || return 1
+ wait $open_pid || return 3
+
+ $CHECKSTAT -t file $DIR/$tdir-1/f || return 4
+ $CHECKSTAT -t file $DIR/$tdir-2/f || return 5
+ rm -rf $DIR/$tdir-*
+}
+run_test 21b "drop open request while close and open are both in flight"
+
+test_21c() {
+ mkdir -p $DIR/$tdir-1
+ mkdir -p $DIR/$tdir-2
+ multiop $DIR/$tdir-1/f O_c &
+ close_pid=$!
+
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000107"
+ mcreate $DIR/$tdir-2/f &
+ open_pid=$!
+ sleep 3
+ do_facet mds "sysctl -w lustre.fail_loc=0"
+
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000115"
+ kill -USR1 $close_pid
+ cancel_lru_locks MDC # force the close
+ wait $close_pid || return 1
+ wait $open_pid || return 2
+
+ do_facet mds "sysctl -w lustre.fail_loc=0"
+
+ $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
+ $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
+ rm -rf $DIR/$tdir-*
+}
+run_test 21c "drop both request while close and open are both in flight"
+
+test_21d() {
+ mkdir -p $DIR/$tdir-1
+ mkdir -p $DIR/$tdir-2
+ multiop $DIR/$tdir-1/f O_c &
+ pid=$!
+
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000129"
+ multiop $DIR/$tdir-2/f Oc &
+ sleep 1
+ do_facet mds "sysctl -w lustre.fail_loc=0"
+
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000122"
+ kill -USR1 $pid
+ cancel_lru_locks MDC # force the close
+ wait $pid || return 1
+ do_facet mds "sysctl -w lustre.fail_loc=0"
+
+ $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
+ $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
+
+ rm -rf $DIR/$tdir-*
+}
+run_test 21d "drop close reply while close and open are both in flight"
+
+test_21e() {
+ mkdir -p $DIR/$tdir-1
+ mkdir -p $DIR/$tdir-2
+ multiop $DIR/$tdir-1/f O_c &
+ pid=$!
+
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000119"
+ touch $DIR/$tdir-2/f &
+ sleep 1
+ do_facet mds "sysctl -w lustre.fail_loc=0"
+
+ kill -USR1 $pid
+ cancel_lru_locks MDC # force the close
+ wait $pid || return 1
+
+ sleep $TIMEOUT
+ $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
+ $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
+ rm -rf $DIR/$tdir-*
+}
+run_test 21e "drop open reply while close and open are both in flight"
+
+test_21f() {
+ mkdir -p $DIR/$tdir-1
+ mkdir -p $DIR/$tdir-2
+ multiop $DIR/$tdir-1/f O_c &
+ pid=$!
+
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000119"
+ touch $DIR/$tdir-2/f &
+ sleep 1
+ do_facet mds "sysctl -w lustre.fail_loc=0"
+
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000122"
+ kill -USR1 $pid
+ cancel_lru_locks MDC # force the close
+ wait $pid || return 1
+ do_facet mds "sysctl -w lustre.fail_loc=0"
+
+ $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
+ $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
+ rm -rf $DIR/$tdir-*
+}
+run_test 21f "drop both reply while close and open are both in flight"
+
+test_21g() {
+ mkdir -p $DIR/$tdir-1
+ mkdir -p $DIR/$tdir-2
+ multiop $DIR/$tdir-1/f O_c &
+ pid=$!
+
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000119"
+ touch $DIR/$tdir-2/f &
+ sleep 1
+ do_facet mds "sysctl -w lustre.fail_loc=0"
+
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000115"
+ kill -USR1 $pid
+ cancel_lru_locks MDC # force the close
+ wait $pid || return 1
+ do_facet mds "sysctl -w lustre.fail_loc=0"
+
+ $CHECKSTAT -t file $DIR/$tdir-1/f || return 2
+ $CHECKSTAT -t file $DIR/$tdir-2/f || return 3
+ rm -rf $DIR/$tdir-*
+}
+run_test 21g "drop open reply and close request while close and open are both in flight"
+
+test_21h() {
+ mkdir -p $DIR/$tdir-1
+ mkdir -p $DIR/$tdir-2
+ multiop $DIR/$tdir-1/f O_c &
+ pid=$!
+
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000107"
+ touch $DIR/$tdir-2/f &
+ touch_pid=$!
+ sleep 1
+ do_facet mds "sysctl -w lustre.fail_loc=0"
+
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000122"
+ cancel_lru_locks MDC # force the close
+ kill -USR1 $pid
+ wait $pid || return 1
+ do_facet mds "sysctl -w lustre.fail_loc=0"
+
+ wait $touch_pid || return 2
+
+ $CHECKSTAT -t file $DIR/$tdir-1/f || return 3
+ $CHECKSTAT -t file $DIR/$tdir-2/f || return 4
+ rm -rf $DIR/$tdir-*
+}
+run_test 21h "drop open request and close reply while close and open are both in flight"
+
+# bug 3462 - multiple MDC requests
+test_22() {
+ f1=$DIR/${tfile}-1
+ f2=$DIR/${tfile}-2
+
+ do_facet mds "sysctl -w lustre.fail_loc=0x80000115"
+ multiop $f2 Oc &
+ close_pid=$!
+
+ sleep 1
+ multiop $f1 msu || return 1
+
+ cancel_lru_locks MDC # force the close
+ do_facet mds "sysctl -w lustre.fail_loc=0"
+
+ wait $close_pid || return 2
+ rm -rf $f2 || return 4
+}
+run_test 22 "drop close request and do mknod"
+
+test_23() { #b=4561
+ multiop $DIR/$tfile O_c &
+ pid=$!
+ # give a chance for open
+ sleep 5
+
+ # try the close
+ drop_request "kill -USR1 $pid"
+
+ fail mds
+ wait $pid || return 1
+ return 0
+}
+#run_test 23 "client hang when close a file after mds crash"
+
+
$CLEANUP